Abstract
Iron ore serves as the fundamental feedstock for blast furnace operations, and its quality is constitutionally affected by the temperature of the mixture during the sintering process. To enhance the early prediction and regulation of the mixture temperature, this study proposes an intelligent 3D prediction model for iron ore temperature based on the Temporal Fusion Transformer (TFT). This model effectively combines advanced multi-horizon forecasting capabilities with temporal dynamic interpretability, while expanding the observation framework into a three-dimensional space through simulation outcomes. Simultaneously, the study focuses on the fluctuation patterns of the major chemical components in sintering materials and their influence on iron ore temperature through the Variational Autoencoder-Temporal Convolutional Networks (VAE-TCN) model. The TFT model, developed using historical sintering data, achieves an [Formula: see text] and RMSE = 4.7568 for one-step-ahead prediction of the sinter temperature spatial distribution, based on a dataset split of 90% training, 5% validation, and 5% testing. Compared with Transformer and Long Short-Term Memory (LSTM) networks, the TFT model demonstrates superior performance, reducing RMSE by 0.805 and 2.9937, respectively. In practical applications, the TFT model offers valuable guidance for real-time temperature monitoring during iron ore sintering operations.